Skip to content

gorged flags

gorged reads its flags with Go's standard flag package (server/gorged, serveFlags). It has no config file and no environment variables of its own. gorged -h prints the same list with the bot-policy options added. Single and double dashes both work (-addr or --addr), and a boolean is switched off with -name=false.

Defaults worth knowing:

  • -addr defaults to :8080, every interface, and that stays the default. Examples here use 127.0.0.1:8090 to keep a trial server on loopback; pick any free port (8090-8099).
  • -dir defaults to gorged-data. Give each server its own directory, such as /tmp/gorge-try. Two servers sharing one corrupt each other (Troubleshooting).
  • -prewarm is on, so a fresh server fetches card art for every dealt deck at start. Pass -prewarm=false offline.

The table below is generated from the flag set of the cmd/gorged binary. TestFlagsDocIsCurrent fails when it differs, so it cannot fall behind. A binary that links a trained-network bot adds a -<policy>-net family of flags for it. make deploy-demo serves the demo (deploy/demo.env.example). The guides explain the flags in use: Play in the browser, Draft events, Serving dirac.

scripts/deploy-demo.sh can serve a binary other than cmd/gorged. Set SERVER_MAIN in deploy/demo.env to a directory holding a main.go that wraps gorged.Main; the script builds it against this checkout and serves it, passing SERVER_EXTRA_ARGS on every server's command line and filling card art for SERVER_EXTRA_DECKS first. scripts/deploy-demo.sh --dry-run prints the resolved configuration, the build commands and each server's command line without building or starting anything.

Flag Default Meaning
-addr :8080 listen address
-allow-decklist-bots false let the server's own bot-only tables run a bot policy that reads every seat's real decklist; such a policy never seats against a human
-art-dir none durable card-art cache directory (default /art, which deploy scripts wipe with the persistence dir)
-auto-mana true enable automatic mana payment plans and controls for human seats (disable with -auto-mana=false for the legacy manual path)
-bot-auto-mana true have hosted bots and human-seat caretakers cast through offered automatic mana payment plans
-bot-common-sense false wrap hosted bots and human-seat caretakers in the bots/sanity CommonSense guard (drop hostile self-removal from the option list, answer the London mulligan by the constructed policy); opt-in, default off
-bot-config none read a bot policy's options from a JSON object file: = (repeatable; overrides built-in defaults, below -bot-opt and the environment)
-bot-opt none set a bot policy option: .= (repeatable; overrides -bot-config and the environment)
-bot-policies none comma-separated set of bot policies this server OFFERS (the /api/bot-policies listing and accepted POST /api/games names); empty offers every registered entry that does not read opponents' decklists (a decklist-reading policy can never be offered: vs-bot always seats a human)
-bot-policy lethal-pressure default hosted bot policy for a vs-bot request that omits bot_policy; must be registered, offered, and support both formats (registered: az-redeal, az-redeal-maxn, bot, cast-profile, forge-heuristic, forge-heuristic-hold, lethal-pressure, sb-first, sb-heuristic, sb-search-lite-atk, sb-tactical, sb-uniform, search)
-bot-search-parallelism 1 goroutines within one searched decision (latency only, never an answer; >= 1)
-bot-search-slots max(1, GOMAXPROCS/2) maximum concurrent searched bot decisions across the process; decisions queue, never degrade (>= 1; default max(1, GOMAXPROCS/2))
-cards .cards corpus directory (ir.gob.gz / cardsfolder)
-commander-ratings none card value file for Commander-draft events ({"Card": {"": value}}, the draftsim -ratings-file format): bots draft and build by it and draft advice ranks by it; empty uses the built-in heuristic
-commander-ratings-field blended which field of -commander-ratings to rate by
-commander-seats 0 seats per commander table; 0 = -seats
-cooldown 5s pause between matches on a perpetual table
-decks directory testkit/testutil/decks directory of deck JSON files; with -engine dirac the default is the first -dirac-sets set's bench decks
-dir gorged-data persistence directory
-dirac-sets fdn with -engine dirac: comma-separated card sets, tried in order for each game (built by make dirac-tables)
-engine gorge the engine this server's tables and vs-bot games run on; one this binary does not link fails at startup, naming those it does
-events false host 4-seat draft events under /api/events (eventhost; persistence under /events)
-events-bot-policy none default game policy of draft-event bot seats that name none; empty uses -bot-policy. Must be registered and offered to events (not a decklist reader), so it may be constructed-only
-feedback feedback directory player bug reports are written to (kept OUT of -dir, which deploys wipe)
-format constructed comma-separated table formats (constructed, commander); table i uses formats[i-1 mod n], e.g. -format commander,constructed runs one of each
-humans none comma-separated slots of table t1 that are real people (e.g. 0,2); t2..tN stay bot tables
-manabrew false mount the ManaBrew protocol adapter (server/host/manabrewhttp) on the main listener; requires -humans or -vsbot
-manabrew-addr none serve the ManaBrew routes on a separate listen address instead of the main one (implies -manabrew)
-manabrew-think 0s override the human-seat think timeout while -manabrew is on (0 = inherit the server default)
-max-events 4 maximum draft events not yet COMPLETE
-max-on-demand-tables 32 maximum retained private play-vs-bot tables per process (0 = unlimited)
-max-search-tables 8 maximum live on-demand tables whose policy is a search entry (0 = unlimited)
-mulligans 7 London mulligans per player before turn 1; 0 disables the pre-game round
-pace 50ms sleep after every decision; 0 = as fast as possible
-paired-bot none let a human player ask a bot to evaluate the decision in front of them: az-closed, or the name of a closed-list search policy. It only advises, and it never sees hidden cards or an opponent's deck list. Off when empty or off
-paired-bot-decks none with -paired-bot az-closed: directory of the decks the bot guesses an opponent's list from (default -decks)
-paired-bot-searches 2 with -paired-bot: maximum paired-bot searches at once across the server; further requests wait (>= 1)
-paired-bot-shared false with -paired-bot: also offer it at a table with more than one human (t1 under -humans), where every seat is told it is on
-paired-bot-sims 100 with -paired-bot az-closed: simulations per request
-perpetual true start a new match when one ends
-pod-bot-decks none enable the az-closed-maxn bot policy: the directory of deck JSON files (e.g. draftsim -decks-out) it guesses an opponent's list from. A closed-list max^n search for Commander pods of up to four seats, offered to tables, draft events (-events-bot-policy) and as a paired bot; it never reads an opponent's real list. Off when empty
-pod-bot-sims 64 with -pod-bot-decks: simulations per searched decision of az-closed-maxn (>= 1)
-pprof false serve Go runtime profiles at /debug/pprof/ on the main listener; refused unless -addr is loopback, and anything proxied to that port can reach them
-prewarm true prewarm the card-art cache for every card in every dealt deck at startup (disable with -prewarm=false)
-prewarm-art-budget 0s with -prewarm-art-only: stop the fill after this much wall-clock time and exit non-zero (0 = no limit)
-prewarm-art-max-consecutive-failures 0 with -prewarm-art-only: stop the fill once this many names in a row have failed and exit non-zero (0 = no limit)
-prewarm-art-only false fill the card-art cache from -decks (into -art-dir) and exit; no server is started. Exits non-zero if any name failed
-sb-agent none register a SpellBench v2 stdio agent as a bot policy: name=/path/to/agent (repeatable; name matches ^[a-z0-9-]+$)
-sbv2 false mount the SpellBench v2 human-seat HTTP routes (server/host/sbv2http) on the main listener; requires -humans or -vsbot
-seat-token none fixed bearer token for the first human slot (tests and local use only; default mints a random token per slot)
-seats 4 seats per table
-seed 1 seed of table 1; table i uses seed+i-1
-spectator public spectator visibility: public (default, every hand hidden) or omniscient (every hand visible)
-tables 4 number of tables
-vsbot false arm the on-demand play-vs-bot flow (landing page seats a human against a bot via POST /api/games)
-vsbot-spectator public spectator visibility of on-demand play-vs-bot tables: public or omniscient